Lab status.
This page was published before the experiment it describes. The result will land here, whatever it turns out to be.
Phase 2: what is being measured
For a stablecoin position in a DeFi protocol: is it better to delegate it to the protocol, or to manage it directly?
The protocols observed are fixed in the frozen methodology, not on this page.
No capital is involved. There is no wallet and no signed transaction. Everything is reading public data and computing on it.
Dates
How it is held to account
- On 10 November the methodology is frozen and its hash is published in two places: on this page, which will be updated that day, and in the public anchors repository, github.com/MontanaMiramar/rall-attestations. Anyone can check later that the methodology was not changed.
- Every day, the record of predictions is anchored to the Bitcoin blockchain with OpenTimestamps. The proofs are public in the same repository, so that, if the record is published, editing any prediction recorded before a day's anchor would show as a mismatch with that anchor. (corrected on 2026-10-29)
- The evaluation runs once, on 11 December at 00:00 UTC, with a script sealed before observation begins.
- Step-by-step instructions for checking the hash and the proofs yourself are published at rodagentic.com/verification/. (corrected on 2026-10-29)
The commitment
On 10 November this page will show the hash of the frozen methodology. After 11 December it will show the result, whatever its sign — as the result of the previous phase was, and that one was negative.
We do not know what Phase 2 will show. That is the point of fixing everything before it starts.
Phase 1: closed
RALL (Rod Agentic Lab Loop), the first FLP node run in the open, ran from 2026-07-25 to 2026-09-22 on a single server, with every action written to an append-only audit log. It showed that a small, gated, auditable agent loop can run unattended for two months. It did not show demand: no external agent ever completed an FLP handshake.
Kept running after the close, as announced: security monitoring and the prompt-injection canary. Paused: the automated daily cycle and outward social activity.
Phase 1 interfaces
The chat for humans (Renata-host) is temporarily unlinked from this site while its review is finished. The MCP endpoint for agents remains available.
MCP endpoint
Model Context Protocol (JSON-RPC 2.0) for autonomous agents.
initialize · tools/list · tools/call · resources/list · resources/read
For the protocol, the lab and Phase 2, the pages on this site are the reference; for anything else, use the contact email below.
Corrections
This page was published on 2026-10-26, before the methodology freeze, as part of the pre-registration. Any change to what it states is recorded here, with its earlier wording.
2026-10-29
- Before: “The proofs are public in the same repository, so no prediction can be written or edited after the fact.” Now: “The proofs are public in the same repository, so that, if the record is published, editing any prediction recorded before a day's anchor would show as a mismatch with that anchor.” Why: the earlier sentence promised more than the proofs support: an anchor does not show when a prediction was written, and an edit can only be detected if the record is published. This was found when /verification/ was published: its list of what cannot be verified independently made the contradiction visible.
- Before: “Step-by-step instructions for checking the hash and the proofs yourself will be published at rodagentic.com/verification/.” Now: “… are published at rodagentic.com/verification/.” Why: that page was published on 2026-10-29.
Contact
For research, collaboration or endorsement: roderickchicas@gmail.com
The protocol itself is open: github.com/MontanaMiramar/flower-of-life-protocol